Hi, On Tue, Jul 09, 2013 at 02:43:45PM -0700, Sarah Sharp wrote: > On Tue, Jul 09, 2013 at 02:59:38PM +0300, Felipe Balbi wrote: > > Hi, > > > > On Mon, Jul 08, 2013 at 09:55:15PM -0700, Sarah Sharp wrote: > > > Felipe, Andy, and Seb, I have a couple questions below. > > > > > > On Fri, Jul 05, 2013 at 08:24:56PM +0300, Xenia Ragiadakou wrote: > > > Felipe, Andy, is there any chance that a platform_device dma_mask > > > pointer would already be initialized by the time the probe function is > > > called? We wouldn't want to overwrite it. Can you please check the > > > xhci_plat_probe code? > > > > yes there is. At least if you're booting with Devicetree, OF core sets > > *all* dma_masks to 32-bits (erroneously IMO): > > > > $ git grep -e dma_mask drivers/of/ > > drivers/of/platform.c: dev->dev.dma_mask = &dev->archdata.dma_mask; > > drivers/of/platform.c: dev->archdata.dma_mask = 0xffffffffUL; > > drivers/of/platform.c: dev->dev.coherent_dma_mask = DMA_BIT_MASK(32); > > drivers/of/platform.c: dev->dev.coherent_dma_mask = ~0; > > drivers/of/platform.c: dev->dma_mask = ~0; > > Ok, so that means Xenia needs to make sure to check whether the dma_mask > pointer is non-NULL in xhci-plat.c, so that she doesn't overwrite it with > a pointer to the coherent dma mask, correct? > > Or should she just overwrite the pointer because the OF core really > shouldn't be setting the DMA mask pointer?
I think it's safe to overwrite (and perhaps even better to do so) provided OF core enables 32-bits blindly, without checking anything about the device. Eventually that dma_mask setting should, likely, come via DeviceTree to the platform_devices. Until, we can overwrite with whatever we can discover out of the device's registers. -- balbi
signature.asc
Description: Digital signature
